Summary of Midterm Evaluations

  1. Grading weekly assignments:

If you have questions about your assignment grades and/or want to discuss how you can improve on your assignments, make an appointment to talk to Alex.

  1. Readings:
    • Full(er) articles
    • Spend more time in class on the results of the articles
    • Boring to just revise the readings
  2. Do we really need to spend so much time on methods?
    • (Good) social science is hard, and the difference (often) depends on the methods
    • You can look up the results online, but you can’t look up how to think about the results.
  3. Group work, but also individual work in class
  4. Class slides as a pdf?
  5. I’ll try to talk more slowly :-P

Learning Goals:

  1. Explain the advantages of field experiments over surveys as tools for documenting the extent of discrimination
  2. Describe the difference between “traditional” vs “aversive racism”
  3. Recognize and apply the concepts of statistical and taste-based discrimination

Motivation

In January 2022, the unemployment rate for all German residents was 6,3 percent. For non-German citizens, it was 13,1 percent.

Motivating question: could these gaps in employment be due to (illegal) hiring discrimination? Or might they be due to something else?

Measuring Discrimination

  1. Why don’t we run a survey and ask people if they’ve been discriminated against in their job search?
  2. What about asking employers how / if they take race or ethnicity into account in their hiring decisions?
  3. What are audit and correspondence tests? Think, then pair-share:
    • How do audit and correspondence tests work?
    • Why are these methods considered the “gold standard” for measuring discrimination?
    • What are the advantages and disadvantages of each approach relative to the other?
    • Is it necessary to send the “matched” applications / testers to the same employer?

Why do Employers Discriminate?

Traditional vs. Aversive Racism

An example of traditional racism can be measured in LaPiere’s 1930 study. When asked “Will you accept members of the Chinese race as guests in your establishment?”, more than 90 percent of the proprietors indicated unequivocal refusal.

Thankfully, much has changed since LaPiere’s time. To quote from Pager and Weston:

“Many individuals in contemporary society experience few conscious anti-Black sentiments, and traditional measures of prejudice have substantially declined. At the same time, there remains a high level of generalized anxiety or discomfort with Blacks that can [unconsciously] shape interracial interaction and decision-making…Aversive racists believe in equality and consciously eschew distinctions on the basis of race; unconscious bias, however, leads to situations in which subtle forms of discrimination persist without the actor’s awareness.”

How might this distinction between (overt and conscious) “traditional” racism and (subtle and unconscious) “aversive” racism explain the discrepancy between what employers say and what they do?

The IAT as a solution?



Taste-based vs. Statistical Discrimination

Definitions


Taste-based discrimination arises when individuals harbour an idiosyncratic preference against interacting with minorities. It is “irrational” in that the employer has no reason to discriminate against the minority candidate, except that he simply dislikes minorities.

In contrast, statistical discrimination emphasizes the ways in which discrimination is rationally-motivated, and focuses on the incomplete information problem facing potential employers.

For instance, employers lacking information on the productivity or “soft skills” of job applicants may use group-level estimates contained in ethnic or racial stereotypes as a screening tool in their hiring decisions.

Importantly, discriminatory behaviour in such models is not driven by a biased response to race or ethnicity per se, but rather by a lack of information about an individual’s true characteristics. In other words, if employers could observe the “true” productivity of every individual applicant, then – under this reasoning – employers would never “need” to discriminate.

BTW: one argument for why correspondence tests / audits generally find low employment discrimination in Germany (compared to other countries) is that German job applications are very detailed. Thus there may be less room for employers to statistically discriminate.


Example of How Studies Usually Test for Statistical Discrimination


Imagine we conduct a correspondence study on housing discrimination.

Suppose that:

  • landlords are only concerned about tenants’ ability to pay the rent.
  • minority renters really are, on average, poorer than majority renters.

We run an experiment where we send a short email to landlords:

Hello,

I saw your flat listed on rentalflats.com and I would like to come by for a showing. Please get in touch with me about some available dates / times.

Regards,

[NAME]

We have 4 signatures, randomly assigned:

  1. Michael Fischer
  2. Mohammed el-Fatih
  3. Prof. Dr. Michael Fischer
  4. Prof. Dr. Mohammed el-Fatih

And we measure the number of responses received by each name.


Questions


Suppose we find discrimination against Mohammed el-Fatih (vs. Michael Fischer), but no discrimination against Prof. Dr. Mohammed el-Fatih (vs. Prof. Dr. Michael Fischer).

What would you infer from this result about why employers discriminate?

Suppose instead we find discrimination against Mohammed el-Fatih (vs. Michael Fischer), and the same amount of discrimination against Prof. Dr. Mohammed el-Fatih (vs. Prof. Dr. Michael Fischer).

What would you now infer about why employers discriminate?


Try it Yourself:


Split up into groups:

In your groups:

  1. Think about a type of job where you could conduct a correspondence test (sending out resumes electronically).
  2. Decide amongst yourselves: what is one reason that employers in this job may statistically discriminate?
  3. Design a simple experiment with 4 treatments that will allow you to test whether this type of statistical discrimination actually exists.


Koopmans et al: Disentangling Taste from Statistics

Koopmans et al. adopt a different approach to disentangling taste-based vs. statistical discrimination.

They first measure the average educational level (education = productivity) and the average cultural values (greater value distance = “distaste”) for a wide range of groups in Germany.

They then conduct a correspondence test:


Finally, they conduct a version of the following regression analysis:

(1) Callback = EthnicName

(2) Callback = EthnicName + education + values

And ask whether the inclusion of measures for education and / or values makes the signifiance of EthnicName go away.

The idea behind this is to ask: holding constant education / values, does Ethnicname still elicit discrimination?

If not, then the “effect” of education / values is interpreted as the reason that Ethnicname is discriminated against in the first place (i.e. education / values are mediating variables).


Questions

  1. What do you think about the decision to use the average education of the group as a proxy for productivity?
    • HINT: think about what types of jobs the researchers applied to.
  2. Given that the applications also included vocational training certificates, letters of reference from previous employers, and diplomas, in what ways might the average education level of the group might still matter?
  3. Would you interpret the value distance results as capturing solely distaste? Or can cultural values also capture aspects of productivity?